Learning Probabilistic Relational Models with Structural Uncertainty

摘要: Most real-world data is stored in relational form. In contrast, most statistical learning methods, e.g., Bayesian network learning, work only with “flat” representations, forcing us to convert our into a form that loses much of the structure. The recently introduced framework probabilistic models(PRMs) allow represent richer dependency structures, involving multiple entities and relations between them; they properties an entity depend probabilistically on related entities. Friedman et al. showed how learn PRMs model attribute uncertainty data, presented techniques for both parameters structure attributes model. this work, we propose methods handling structural PRMs. Structural over which are domain. We two mechanisms modeling uncertainty: reference uncertaintyand existence uncertainty. describe appropriate conditions using each present algorithms each. conclude some preliminary experimental results comparing contrasting use these mechanism domains